Buyers are already asking ChatGPT and Gemini where to buy in Dubai, whether off-plan is safe, and which developer to trust. The party with the most riding on the answer - the developer - is usually the one missing from it.
The query set isn't exotic. It's the same handful of questions a serious buyer has always asked a broker, now typed into ChatGPT first: "best area to buy property in Dubai for families," "best area for rental yield," "is off-plan safe in Dubai," "golden visa property threshold," "is [developer] reliable," and "what's the rental yield in [area]." Each of these has an intent behind it - area selection, risk assessment, developer trust, or investment return - and each one is now answered by a machine before a human agent gets the call.
Run these questions and a pattern holds across engines. Three source types dominate:
Developer sites show up occasionally, and almost always as a single named project mentioned inside someone else's comparison, not as the source doing the comparing.
It's structural, not a visibility bug. A developer's website exists to sell the current launch: hero renders, a payment plan graphic, a brochure PDF behind a lead form. That's a page built for a human who has already decided to look at that project. It's not built to answer the questions a buyer asks before deciding - which area, which developer, is off-plan even safe.
Engines need facts they can cite without guessing: yield by unit type, service charge history, delivery track record against the original handover date, how this payment plan compares to the last one. Portals and agencies publish exactly that, in tables, with dates attached. Most developer sites publish none of it, and what they do publish is often behind JavaScript-heavy renders that are slow or awkward to crawl in the first place.
"Is off-plan safe" isn't a question a developer can dodge by staying quiet. It gets asked and answered whether or not the developer participates - through escrow rules, RERA registration under Law 13 of 2008, and general track-record commentary drawn from news coverage and agency writeups. A buyer asking "is [developer] reliable" gets an answer built from whatever third parties have published about that developer's delivery history. If the developer has never published its own handover record, the engine still finds one, just not from the developer.
When the AI answer flow runs area first, then portal, the developer meets the buyer last - after the buyer already has a shortlist, a yield expectation and a portal relationship. That's a worse negotiating position and it shows up in margin, not just in traffic. The intermediary that answered the early questions keeps the relationship; the developer becomes one line in someone else's comparison table.
We run the actual buyer questions - area choice, off-plan safety, developer trust, yield by area - against ChatGPT, Claude, Gemini and Perplexity, and read what comes back: which sources get cited, which developers get named, and where the answer sends the buyer instead. That's the same diagnostic approach we use across categories, covered in what AI discovery intelligence actually measures. The pattern of getting cited next to a competitor, or not at all, isn't unique to real estate - see why ChatGPT recommends your competitor for the general mechanism. For a developer, the fix starts with publishing the facts a buyer would need to compare projects, in a format an engine can actually read and cite.